精神病医疗记录中的机器学习:创伤注释的黄金标准方法
Eben Holderness1,2, Bruce Atwood1, Marc Verhagen2
1Psychosis Neurobiology Laboratory, McLean Hospital, Belmont, MA, USA.
Translational psychiatry
|August 1, 2025
概括
研究人员从精神病医疗记录创建了一个黄金标准数据集,用于训练机器学习模型. 这一数据集有助于检测症状,物质使用和创伤,推进精神病医疗保健.
科学领域:
- 精神病学是一个精神病学.
- 计算语言学 计算语言学
- 机器学习 机器学习
背景情况:
- 精神病学电子健康记录 (EHR) 由于其非结构化的性质,对机器学习具有挑战性.
- 开发用于精神病治疗的机器学习模型需要高质量的注释数据集.
研究的目的:
- 为患有精神障碍和创伤后应激障碍 (PTSD) 的患者创建注释精神病学EHR的黄金标准数据集.
- 制定临床信息的指导方针,用于注释创伤事件.
- 为了证明数据集在训练机器学习模型中的实用性,用于症状,物质使用和创伤检测.
主要方法:
- 编制了200个叙事重重的精神病学EHR的语料库.
- 与临床专家和计算语言学家一起开发了一个详细的注释方案.
- 对创伤相关事件和临床信息进行注释的EHR,实现了高的注释者间一致性 (跨度为0.715,属性为0.874).
主要成果:
- 建立了第一个黄金标准数据集,用于标记精神病EHR中的创伤特征.
- 机器学习模型实现了高性能 (跨度为0.76的微F1,属性为0.82).
- 高度的注释者间协议和模型性能证明了数据集的可靠性.
结论:
- 创建的黄金标准数据集适合在精神病医疗保健中训练机器学习模型.
- 本资源促进了机器学习的应用,以了解疾病异质性和治疗影响.
- 该数据集有助于更好地检测精神病患者的症状,物质使用和创伤.
相关概念视频
Post-traumatic Stress Disorder
114
Post-traumatic stress disorder (PTSD) is a psychiatric condition that arises following exposure to traumatic events such as natural disasters, forced displacement, or severe accidents. It significantly impairs individuals' ability to cope with daily activities and disrupts their emotional and psychological equilibrium.
Symptoms and Behavioral Manifestations
A spectrum of distressing symptoms characterizes PTSD. Recurrent flashbacks, where individuals involuntarily relive traumatic events,...
Symptoms and Behavioral Manifestations
A spectrum of distressing symptoms characterizes PTSD. Recurrent flashbacks, where individuals involuntarily relive traumatic events,...
114
Traumatic Memory
181
Emotionally traumatic events often lead to memories that are exceptionally vivid and enduring, sometimes persisting with remarkable clarity throughout an individual's life. A classic example of this phenomenon is a person who survives a car accident. Even years later, they may recall every detail of the event with startling accuracy — the screeching of the tires, the jarring impact, and the acrid smell of burning rubber. Such vividness contrasts sharply with how an individual...
181


